Formulation of the transient ensemble experiments

Now that we have finished running one 5-member ensembles from 1950 to 1990, it is time to consider more exactly how the experiments should be carried out. The general idea is to choose initial conditions from our transient ensemble and run out those as new ensembles. The model formulation for the experiment ensembles will be the same as for the initial condition ensembles with the exception that volcanic eruptions will not be included in the former (as eruptions are not possible to forecast). The size of the experiment ensembles will vary depending on the nature of the experiment and how heavily the cluster is being used. I would suggest ensemble sizes between 8-20.

In our initial condition ensembles we are hoping to find large, but naturally occuring, anomalies. I suggest that the anomalies are defined as differences from the all ensemble mean 1950-1980. Click on the link at the end of this sentence to discuss key diagnostics. One thing to keep in mind is that these are transient ensembles, so not only the initial conditions need to be considered, but also the changing external forcing. This is what makes our experiments novel and enables us to answer questions such as "does the impact of GHG forcing on decadal timescales depend on initial conditions?".

Once an anomaly has been identified, an interesting experiment to do is to start an ensemble off at a point 6 months or a year before that anomaly occured. The number of ensemble members who then develop the same anomaly will be an indication of how predictable this anomaly is and how important the initial conditions are. It will also be possible to study the dynamics that lead to the build-up of the anomaly. To get more of an idea of which initial conditions are important in the build-up to the anomaly, it may also be possible to do further experiments using anomaly assimilations, see AssimilationIdeas.

As a continuation of the first experiment, or as an experiment in its own right, the next step would be to start an ensemble from the point where the anomaly is at its maximum. An additional ensemble would also be started which has the same anomaly of the opposite sign or no anomaly at all, but with otherwise similar initial conditions. By comparing the spread of the two ensembles it will be possible to see how much predictability the anomaly gives rise to. It will also be possible to investigate which dynamics in the model act to re-inforce the anomaly and which dynamics act to break it down. Also of interest are the down-stream effects of the anomaly (is it part of an oscillation?) and the impact on climate (surface temperature, precipitation, etc).

-- LeonHermanson - 26 Jun 2007

UPDATE: Unless we are very lucky, we have probably lost the one of the transient ensembles in the great pegasus data loss disaster of '07.